Intel next-generation CPU architecture, code-named Nehalem, and now known as Core i7, will be officially launched in Q4 of this year, most likely in November.
HEXUS is out in force at this year's IDF (Intel Developer Forum), and, as usual, we'll be covering everything of note.
However, arriving in San Francisco a couple of days and, rudely, finding a Nehalem-based sytem at the Moscone Center - IDF's home - where the test systems were being built, HEXUS managed to secretly grab some time with a Nehalem box.
We managed to benchmark the single-socket, quad-core, eight-threaded, 2.93GHz monster and compare it against Intel's very own Core 2 Extreme QX9770, QX6800, and AMD's Phenom 9950 Black Edition.
As a summary, we know that the monolithic, 45nm-based Nehalem microarchitecture has an integrated memory controller - supporting tri-channel DDR3-1,066MHz RAM for the desktop - QuickPath Interconnect, simultaneous multithreading, and a three-level cache hierarchy with a large pool shared amongst all cores. The basic design, however, is entrenched in Core 2 roots, but the enhancements bear remarkable similarities to what AMD has been doing for a while now, or are we just being cynical and myopic?
Nehalem will be available for servers, desktops, and mobiles, and it's reasonable to assume that Intel will disseminate further model-numbering details during the Forum this week.
Intel's briefing documents indicate that initial single-package Nehalems will ship with a quad-core arrangement, bearing eight threads with SMT, with octo-core (16-thread) models to follow.
Bloomfield, the desktop part, will be available with frequencies of 2.66GHz, 2.93GHz, and 3.2GHz, priced, we believe, at $266, $532, and $999, respectively, upon launch. Of course, a new motherboard will be needed to house the LGA1366 processors, and we saw plenty of Intel X58-based boards on show at this year's COMPUTEX event.
Prima facie, Nehalem - or Core i7 - is everything Core 2 is but more, lots more, and that's troubling for AMD, who's currently struggling with ramping clock speeds on its quad-core Phenom CPUs, thereby, with due note to Intel's line-up, pigeon-holing Phenom pricing to sub-$200.
Here's the Nehalem box that was equipped with a 2.93GHz-clocked CPU. Note the 6 DIMM slots for the X58 pre-production board. Three DIMM slots were filled, for tri-channel operation, using Kingston DDR3-1,066MHz modules.
Knowing that Nehalem doesn't use a front-side bus, the overall clockspeed is calculated by a multiplier that's factored by a 133MHz 'bus' speed.
The test system ran with the QuickPath interconnect operating via a 20-bit-wide bus and with a 4.8GT (gigatransfers) speed, pushing around 24GB/s for the link.
The tri-channel memory, making a total of 192 bits, offers around 38.4GB/s when using DDR3-1,600 memory - which is unoffically supported right now - and even 25.6GB/s with DDR3-1,066MHz RAM, which fits nicely inside the QuickPath link and provides plenty more than current Intel offerings.
Vista has a hard time working out what it is, though...
We had our labs build near-identical comparison systems, as far as other components were concerned, and have benchmarked Intel's leading single-CPU SKU, the Core 2 Extreme QX9770. Numbers from the 65nm-based Core 2 Extreme QX6800 have been included, to serve as an indicator of clock-for-clock performance: the Nehalem processor operates at 2.93GHz, too.
We ran an nForce 790i Ultra SLI for the two Intel LGA775 CPUs, as the table shows, for the simple but annoying reason that our X48 platform would refuse to run DDR3-1600MHz memory with utter stability.
Lastly, an AMD Phenom X4 9950 Black Edition has been included in to the mix, to see how much of a mountain AMD has to climb in order to gain performance parity with what will be Intel's second-rung Nehalem.
The test system included slower-than-expected DDR3-1,066MHz memory in tri-channel mode and, as you will see when we arrive at the gaming benchmarks, there was some untoward graphics-related behaviour exhibited, too.
Please, please bear in mind that this is a look at a pre-production CPU and motherboard, and final numbers may not be indicative of retail performance. It's also not an apples-to-apples comparison, insofar as value is concerned.
Hello! The removal of the archaic front-side bus and use of tri-channel DDR3 memory, albeit at 1,066MHz for the Nehalem system, opens up the way to some incredible memory-bandwidth results - the highest we've seen for a consumer setup.
Just like graphics, memory bandwidth plays a pivotal role in determining overall performance, so Nehalem's memory interface and onboard controller look good on first glance.
Latency is average, beaten out by the high-bandwidth memory on the LGA775 systems. We're adamant that Nehalem, at the very least, will provide a Core 2 Extreme-matching figure when we officially review the processor in Q4 of this year.
HEXUS' PiFast test calculates the constant Pi to 10m places, using a brute-strength approach. What's interesting is that the single-threaded test is almost as fast on the 2.93GHz Nehalem as on the 3.2GHz QX9770, suggesting that memory bandwidth is coming into serious play.
AMD's fastest consumer CPU, the 2.6GHz-clocked Phenom X4 9950 BE, is some significant way behind.
On a clock-for-clock basis against the QX6800, the Nehalem is around 10 seconds faster in the dual-threaded WAV-encoding test, but it doesn't quite have the architectural muscle to push the 3.2GHz C2E into second place.
Installing and using the latest iteration of DivX, which takes partial advantage of the eight-thread ability of the Core i7, we see that a C2E QX9770 is some 15 per cent slower, an equivalently-clocked, non-SSE4.1-enabled QX6800 is 35 per cent slower , and a whopping 72 per cent deficit for Phenom X4 9950 BE.
The story is even brighter in WinRAR compression, which shows the next best, QX9770, some 30 per cent slower.
Professional software, which is usually optimised for multi-core usage, shows Nehalem in an appropriately refulgent light; trampling on the C2E QX9770 in unceremonious fashion. Indeed, it's over 33 per cent faster than the aforementioned CPU in Pov-ray 64-bit, and over twice as potent as AMD's best. Scary, huh?
We alluded to the fact that it wasn't all roses in the Nehalem garden, as far as the test box was concerned. One look in Device Manager showed that not all the correct drivers had been installed, which did little to hinder 2D performance, but played a part in sub-optimal 3D results.
A table has been included to highlight the results we observed, but it is abundantly clear that something was awry in the test box.
A few things to note here, folks.
Firstly, the Quake Wars: ET 1,680x1,050 result is significantly lower than expected, because the test becomes practically GPU-limited at that setting, even on a Radeon HD 4870 512MB card: we see that from the similarity of the results between other CPUs.
Secondly, the 3DMark Vantage default test score is around 1,000 marks too low, again resulting from, we believe, an unoptimised setup. But take a look at the CPU-only score and Nehalem's power rears its head.
Nehalem won't matter much if you play games that are limited by the graphics subsystem, usually at higher resolutions and image-quality settings, but it's always nice to have extra power under the hood, we suppose.
We will, of course, be taking an in-depth look as soon as we have a system that's not disadvantaged by time constraints.
What do we make of our first hands-on look with Nehalem - Intel's next-generation processor architecture?
Nehalem's performance is born from taking the Core 2 architecture as a base and adding sensible, performance-enhancing additions such as an integrated memory controller, QuickPath interconnect, tiered cache, and tri-channel memory. Last but not least, SMT (simultaneous multithreading) provides a healthy boost, too.
Looking back through the numbers, the 2.93GHz Nehalem naturally comes into its own when the cores, be they physical or virtual, are pushed by the software. When this happens, it's up to 33 per cent faster than a 3.2GHz Intel Core 2 Extreme QX9770, and some 50 per cent quicker than an equivalently-clocked (Kentsfield-based) Core 2 Quad CPU.
Talking about the consumer space, Intel has enjoyed a performance lead over AMD for some time, and it's only going to continue, unabated, with the release of the Nehalem core. AMD's response will come in the form of a Phenom-upgraded processor, Shanghai, but it will need to be something special, really, really special, for it to effectively counter the Nehalem threat.
Knowing the likely pricing of initial Nehalem CPUs, the ramifications of the new architecture are more than just related to performance, however. AMD, it seems, will not be able to compete against the Nehalem's performance, even with Shanghai, and, so, the pricing of subsequent Phenoms will be indirectly set by Intel. We can't see a Phenom being sold for, say, $500 anytime soon; AMD doesn't have the necessary power to do so, unless it can pull a rabbit out of the silicon bag.
Bottom line: Nehalem's very real performance numbers put incumbent Core 2 Extreme in the shade and pose AMD some really tough headaches.
http://www.hexus.net/content/item.php?item=15015
Core I7 Updates Thread -
20th August 2008
